Fine phonetic labeling methodology for speech recognition research

A methodology is described for producing line phonetic transcriptions of speech for research data bases. The detailed labelings can be used in experiments to analyze performance of components in an acoustic-phonetic based automatic speech recognition system. An ASCII-based symbology and labeling syntax have been developed which describe classes of observable, linguistically-relevant phenomena, as well as acoustically-distinct phenomena. The labeling scheme employs diacritic markers to describe allophonic variations (e.g., devoiced variants, vowel coloring). There are additional descriptors for voice quality (e.g., creak, breathiness).